Project Summary/Abstract Dramatic advances in our understanding of molecular structure and function promise to accelerate the creation of new diagnostics and therapeutics. However the link between the structure of a biological macromolecule and its function is usually not obvious: fundamental to understanding how a molecule functions is an understanding of how its structure behaves over time. Recent advances in molecular dynamics simulations now allow the rapid collection of information about structural motion. These data sets are huge, and require statistical machine learning algorithms to characterize and recognize patterns relevant to function. The National Library of Medicine's new long-range plan calls for research in the use of advanced simulation and machine learning algorithms in support of biomedical research. This proposal focuses on annotating molecular structures with missing or incomplete functional information. We are particularly interested in identifying binding sites and active sites in proteins. We bring together simulation and machine learning, and hypothesize that the performance of structure- based function annotation methods will dramatically improve with the addition of information about dynamics. Thus, our specific aims are (1) to develop methods for recognizing function from structural dynamics and diversity, (2) to develop capabilities for large scale clustering and analysis tools for the discovery of novel functions, and (3) to apply our tools to challenging and important biological systems, while disseminating our software, data and capabilities to the biomedical research community. In particular, we will focus our new capabilities on three difficult function annotation challenges: ATP binding sites, phosphorylation sites, and metabolizing enzyme active sites.